AgentEye

About

Point at what's wrong. Your AI agent fixes it.
                                               
Describing UI bugs in words is painful. "The button — no, the other one — it's 3px off and the color is wrong on hover." Screenshots help, but your AI agent can't parse them into code changes. You end up being the translator between what you see and what the agent needs.                                                                                                                                                        
                                                                                                                                                                                                                                                                                                                                                                                                                                         
AgentEye closes that gap. Click any element, say what needs to change, and copy structured feedback your agent can act on — with the exact CSS selectors, component names, and source file paths it needs to make the edit.                                                                                                                                                                                                            
                                                                                                                                                                                                                                                                                                                                                                                                                                         
HOW IT WORKS                                                                                                                                                                                                                                                                                                                                                                                                                           
                                     
  1. Activate AgentEye on any webpage — local dev, staging, or production
  2. Hover to highlight elements and see their tag, CSS selector, and React or Svelte component tree
  3. Click to drop a numbered pin, then type what's wrong                                                                                                                                                                                                                                                                                                                                                                                
  4. Hit Copy to get clean Markdown with selectors, source file paths, bounding boxes, and your notes
  5. Paste into Claude Code, Cursor, Copilot, or ChatGPT — the agent knows exactly where to edit                                                                                                                                                                                                                                                                                                                                         
                                                                                                                                                                                                                                                                                                                                                                                                                                         
WHY IT WORKS BETTER THAN SCREENSHOTS         
                                                                                                                                                                                                                                                                                                                                                                                                                                         
Framework-aware — Detects React and Svelte components automatically. Shows the component hierarchy and maps elements to source files, so your agent edits the right file instead of guessing.                                                                                                                                                                                                                                          
                                         
Freeze mode — Pause animations, videos, and CSS transitions to annotate states that are normally impossible to capture: hover menus, loading spinners, mid-transition states.                                                                                                                                                                                                                                                          
                                        
Four detail levels — Compact, Standard, Detailed, or Forensic. From a one-line selector to a full report with computed styles, accessibility attributes, viewport size, and DOM path.                                                                                                                                                                                                                                                  
                                
Multi-select — Annotate several elements at once for batch feedback in a single paste.                                                                                                                                                                                                                                                                                                                                                 
                                               
Zero interference — Runs in shadow DOM. Never injects into your page, never affects layout, never breaks your app.                                                                                                                                                                                                                                                                                                                     
                       
BUILT FOR TEXT-BASED AI WORKFLOWS                                                                                                                                                                                                                                                                                                                                                                                                      
                                         
AgentEye outputs structured data, not screenshots. Your agent parses element selectors, component names, and file paths directly. No vision model needed. Works with any AI tool that accepts text.
                       
No account required. All data stays local in your browser.